NIST Generative AI Profile
A companion resource for applying AI risk-management guidance to generative systems.
Reviewed · 2026-10-071 cited source
Skip to Observatory content← AI Risk Registry FOLLOW THE CONNECTIONS INSPECT THE ORIGINAL WORK
AI-system security
Adversarial inputs and vulnerable integrations can undermine an AI application’s intended behavior.
- Last reviewed
- Jurisdiction / scope
- Global
- Record date / period
- Date not established
- Assessment confidence
- Not scored
Overview
Security includes the model, its tools, its data and the surrounding application.
Why it matters
Authority to read data or act on a user’s behalf changes the consequences of failure.
Current evidence
NIST’s GenAI Profile discusses information-security risk. OpenAI’s historical GPT-4 documentation acknowledges adversarial prompts.
What we know
These sources identify concerns; they do not establish a breach in every connected system.
What we do not know
Which attacks remain feasible in a particular deployed configuration?
Key uncertainties
Access boundaries, mitigations, attacker knowledge and evaluation conditions.
Current assessment
Examine the actual permissions and deployment context.
Methodology
Documented risk framing; no invented attack success rates.
Related records.
Links provide context. Read the stated relationship; inclusion does not imply misconduct, endorsement or a legal obligation.
Research
Models
Sources.
- GPT-4 research announcement ↗
OpenAI · Primary source · Developer documentation
Published 2023-03-14 · Accessed 2026-10-07 - Generative Artificial Intelligence Profile (NIST AI 600-1) ↗
NIST · Primary source · Government technical report
Published 2024-07 · Accessed 2026-10-07DOI: 10.6028/NIST.AI.600-1
Record history.
Review dates track editorial checks. Updates may reflect corrections or added context; they are not new incidents. Private draft revisions are omitted.
- · Revision 1
Initial sourced historical collection. Review date describes this record, not the date of the event.